Integrating Speech and Iconic Gestures in a Stroop-like Task: Evidence for Automatic Processing
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
Integrating Speech and Iconic Gestures in a Stroop-like Task: Evidence for Automatic Processing Spencer D. Kelly1, Peter Creigh2, and James Bartolotti1 Abstract ■ Previous research has demonstrated a link between lan- racy rates, RTs, and ERPs were recorded to the words. Although guage and action in the brain. The present study investigates not relevant to the task, participants paid attention to the se- the strength of this neural relationship by focusing on a poten- mantic relationship between the speech and the gesture, pro- tial interface between the two systems: cospeech iconic ges- ducing a larger N400 to words accompanied by incongruent ture. Participants performed a Stroop-like task in which they versus congruent gestures. In addition, RTs were slower to in- watched videos of a man and a woman speaking and gesturing congruent versus congruent gesture–speech stimuli, but this ef- about common actions. The videos differed as to whether the fect was greater when the gender of the gesturer and speaker gender of the speaker and gesturer was the same or different was the same versus different. These results suggest that the and whether the content of the speech and gesture was congru- integration of gesture and speech during language comprehen- ent or incongruent. The task was to identify whether a man or a sion is automatic but also under some degree of neurocognitive woman produced the spoken portion of the videos while accu- control. ■ INTRODUCTION have argued that the two together constitute a single inte- Over the past decade, cognitive neuroscientists have be- grated system (McNeill, 1992, 2005; Kita & Özyürek, 2003; come increasingly interested in the relationship between Özyürek & Kelly, 2007). For example, McNeill (2005) argues language and action in the human brain (for a recent that because speech and iconic gesture temporally overlap review, see Willems & Hagoort, 2007). This interest is but convey information in two very different ways—speech fueled by the discovery that language shares neural sub- is conventionalized and arbitrary, whereas iconic gestures strates that are involved in more basic processes such as are idiosyncratic and imagistic—the two together capture the production and the perception of action (Nishitani, and reflect different aspects of a unitary underlying cognitive Schurmann, Amunts, & Hari, 2005; Rizzolatti & Craighero, process. For example, imagine someone saying, “They were 2004; Rizzolatti & Arbib, 1998). These findings have led up late last night,” while making a drinking gesture. From this many researchers to view language as an ability that grew example, it should be clear that the iconic gesture and the out of action systems in our evolutionary past (Armstrong accompanying speech combine to reveal meaning about & Wilcox, 2007; Corballis, 2003; Kelly et al., 2002), or as an activity that is not fully captured in one modality alone. Elizabeth Bates so eloquently put it, “[as] a new machine Exploring this relationship between speech and iconic that nature has constructed out of old parts” (MacWhinney gesture may provide a unique opportunity to better un- & Bates, 1989; Bates, Benigni, Bretherton, Camaioni, & derstand the extent to which language and action are Volterra, 1979). Taking this embodied view that language connected in the brain. Indeed, if language is inextricably is inextricably tied to action in present day communica- linked to action, as many have argued, one might expect tion, the current article investigates the strength of this iconic gesture to be inextricably linked to speech. The relationship by focusing on a potential interface between present study investigates the strength of this neural link the two systems: cospeech iconic gesture. by exploring the extent to which the two channels are Cospeech iconic gestures are spontaneous hand move- automatically integrated during language comprehension. ments that naturally, pervasively, and unconsciously accom- pany speech and convey visual information about object attributes, spatial relationships, and movements (McNeill, The Integration of Gesture and Speech 2005; Kendon, 2004; Goldin-Meadow, 2003; Clark, 1996). These gestures are so tightly tied to speech that theorists The first attempts to address this question of whether ges- ture and speech form an integrated system came from re- searchers in psychology, linguistics, and education using behavioral methods. Although some researchers have re- 1 Colgate University, 2University of Pittsburgh cently attempted to support this claim by investigating © 2009 Massachusetts Institute of Technology Journal of Cognitive Neuroscience 22:4, pp. 683–694
processes involved in language production (Goldin- gestures had meaningful versus nonmeaningful relation- Meadow, 2003; Kita & Özyürek, 2003), the majority of ex- ships to speech. Moreover, Willems et al. (2007) showed that perimental work has focused on language comprehension, gestural information and spoken information are both pro- showing that people typically pay attention to gesture cessed in the same brain regions—inferior frontal areas— when processing speech (Beattie & Shovelton, 1999; during language comprehension. This inferior frontal region Cassell, McNeill, & McCullough, 1999; Kelly, Barr, Church, is in an interesting site for the integration of gesture and & Lynch, 1999; Kelly & Church, 1998; Goldin-Meadow, speech because it is traditionally seen as a language region Wein, & Chang, 1992). However, these studies have been (as in Brocaʼs area), but more recently it has been described criticized for not conclusively demonstrating that gesture as a key component to the human mirror neuron system and speech are truly integrated during language compre- (Rizzolatti & Craighero, 2004). Other researchers have hension. For example, researchers such as Robert Krauss identified more posterior regions, such as the superior tem- have argued that this work on gesture comprehension de- poral sulcus and the inferior parietal lobule, as sites for monstrates, at best, a trivial relationship between speech the integration of gesture and speech (Holle et al., 2008; and gesture at comprehension (Krauss, 1998; Krauss, Morrel- Wilson et al., 2008). Again, these areas are noteworthy be- Samuels, & Colasante, 1991). That is, gesture may be used as cause they, like the more anterior regions, are independently “add-on” information to comprehend a communicatorʼs implicated in language processing and action processing. meaning only after the speech has been processed. In other This research has made good progress toward showing words, in cases where gesture does influence comprehen- that gesture and speech are indeed integrated during lan- sion, the attention to gesture is post hoc and occurs well guage comprehension and that this integration occurs in after the semantic processing of speech. So according to brain regions involved with human action. Still, it does not this view, when people do pay attention to gesture during fully address the critique by Krauss et al. (1991) regarding language comprehension, it is an afterthought. the extent to which gesture and speech are integrated dur- Recently, researchers have turned to techniques in cog- ing comprehension. That is, gesture may be linked to nitive neuroscience to better understand the relationship speech, but—to borrow an earlier description—not in- between speech and gesture during language comprehen- extricably linked. One way to address this concern is to sion. Many of the initial studies used ERPs to investigate determine whether people cannot help but integrate ges- the time course of gesture–speech integration (Bernardis, ture and speech. Salillas, & Caramelli, 2008; Holle & Gunter, 2007; Özyürek, Willems, Kita, & Hagoort, 2007; Wu & Coulson, 2007a, Automatic and Controlled Processes 2007b; Kelly, Kravitz, & Hopkins, 2004). For example, Wu and Coulson (2007b) presented gesture–speech ut- To determine the strength of the gesture–speech relation- terances followed by pictures that were related either to ship, it may be highly revealing to test whether the integra- gesture and speech or to just the speech alone. When tion of the two modalities is automatic or controlled. This pictures were related to gesture and speech, participants distinction was first introduced within the field of cognitive produced a smaller N400 component (the traditional se- psychology (Schneider & Shiffrin, 1977; Posner & Snyder, mantic integration effect, as in Kutas & Hillyard, 1980) 1975; but for a more recent perspective, see Bargh & than when the pictures were related to just the speech. Morsella, 2008). The traditional distinction is that auto- This suggests that the visuospatial aspects of gestures matic processes are low level, fast, and obligatory, whereas combined with speech to build stronger and more vivid controlled processes are high level, slow, and under inten- expectations of the pictures than just speech alone. This tional guidance. For example, recognizing that someone finding makes sense in light of other ERP research demon- who reaches for an object must want that object is an strating that the semantic processing of a word is affected automatic process—understanding why they want that ob- by the presence of cospeech gestures that convey either ject, a controlled process. So when someone is presented congruent or incongruent visual information (Kelly et al., with gesture and speech, do they automatically integrate 2004). the two modalities or do they more consciously and stra- So where are gesture and speech semantically inte- tegically choose to integrate them? grated in the brain? Using functional neuroimaging tech- Researchers have found evidence in other domains that niques (mostly fMRI), researchers have discovered that certain nonverbal or paralinguistic behaviors (such as this integration occurs in brain regions implicated in facial expression, body posture, and tone of voice) are the production and comprehension of human action automatically processed in the brain (de Gelder, 2006; (Hubbard, Wilson, Callan, & Dapretto, 2009; Holle, Gunter, Winston, Strange, OʼDoherty, & Dolan, 2002; Pourtois, Rüschemeyer, Hennenlotter, & Iacoboni, 2008; Wilson, de Gelder, Vroomen, Rossion, & Crommelinck, 2000). For Molnar-Szakacs, & Iacoboni, 2008; Montgomery, Isenberg, example, de Gelder (2006) provides evidence for a sub- & Haxby, 2007; Skipper, Goldin-Meadow, Nusbaum, & cortical neurobiological detection system (which primarily Small, 2007; Willems, Özyürek, & Hagoort, 2007). For ex- includes the amygdala) that obligatorily and unconsciously ample, Skipper et al. (2007) found that inferior frontal re- processes emotional information conveyed through fa- gions processed speech and gesture differently when the cial expressions and body posture. Focusing specifically 684 Journal of Cognitive Neuroscience Volume 22, Number 4
on manual actions, there is evidence that noncommuni- not requiring any overt or conscious attention to the se- cative hand movements, such as reaches and physical mantics of the stimuli. manipulations of objects, also elicit automatic processing The third point is more general and requires some (Iacoboni et al., 2005; Blakemore & Decety, 2001). For ex- explanation. Previous ERP research on the time course ample, Iacoboni et al. (2005) found that the inferior fron- of integrating speech and cospeech gestures has con- tal gyrus and the premotor cortex (regions traditionally flated iconicity and indexicality (Kelly et al., 2004, 2007).1 associated with the human mirror neuron system) differ- These previous studies used stimuli of people gesturing entiated actions that were intentionally versus unintention- to objects—for example, gesturing the tallness of a glass ally produced. Interestingly, this effect held even when and verbally describing either the tallness of that glass people were not explicitly instructed to attend to the goals (a congruent pair) or the shortness of an adjacent dish of the actions, a finding that led the authors to conclude (an incongruent pair). The main finding was that incon- that people automatically process the meaning of human gruent pairs produced a larger N400 component than action. congruent pairs. However, it is impossible to determine There have been a handful of studies that have specif- whether the N400 effect arose from the incongruent ically explored the automatic versus controlled nature of iconic information of the gesture (i.e., tall vs. short) gesture processing (Holle & Gunter, 2007; Kelly, Ward, or whether it arose from the incongruent indexical in- Creigh, & Bartolotti, 2007; Langton & Bruce, 2000). On formation (i.e., glass vs. dish). It is one thing to say that the one hand, Langton and Bruce (2000) demonstrated indicating one object with a gesture makes it difficult to that people cannot help but pay attention to pointing simultaneously process speech referring to another ob- and emblematic gestures (in the absence of speech) even ject, but it is another to say that processing the iconic when their task was to attend to other information. On the meaning of a gesture interacts with the simultaneous se- other hand, there is some evidence that when it comes to mantic processing of the accompanying speech. If one processing cospeech gestures, attention to gesture may wants to make strong and definitive claims about the have elements of a controlled process. For example, Holle automatic or controlled integration of speech and co- and Gunter (2007) showed that the presence of nonmean- speech gesture in the brain, it is very important to pre- ingful gestures (e.g., grooming behaviors) modulates the cisely identify what aspects of gesture—its indexicality extent to which meaningful gestures set up semantic ex- or its iconicity—are part of that integrated system. pectations for speech later in a sentence. In addition, Kelly et al. (2007) found that the integration of gesture and The Present Study speech was influenced by explicit instructions about the intended relationship between speech and gesture. Spe- The present study used improved methods to explore the cifically, when processing speech in the presence of an automatic integration of gesture and speech during lan- incongruent versus congruent gesture, there was a larger guage comprehension. First, to address the issue of “ico- N400 effect in bilateral frontal sites when people were told nicity versus indexicality,” we used gestures that conveyed that the gesture and speech belonged together, but this no indexical information. The gestures in the present N400 effect was not preset in the left hemisphere region study conveyed iconic information about various actions when people were told that gesture and speech did not in the absence of actual objects. These types of iconic ges- belong together. In other words, people appeared to have tures are very common in natural communication (McNeill, some control over whether they integrated gesture and 1992), but their meaning is usually unclear or ambiguous speech, at least under circumstances of explicit instruc- in the absence of accompanying speech (unlike indexical tions about whether to integrate the two modalities. gestures). Therefore, better understanding how they are This finding from Kelly et al. (2007) is provocative be- temporally integrated with co-occurring speech is impor- cause it suggests that high-level information such as knowl- tant for theories of gesture and speech, specifically, and edge about communicative intent modulates the neural action and language, more generally. integration of gesture and speech during language com- Second, to address the problem of heavy-handed task prehension. However, the study leaves open a number instructions that require conscious analysis of gesture of unresolved issues. First, because it used very direct and speech, we used a Stroop-like paradigm to test the and heavy-handed (as it were) instructions not to integrate integration of the two modalities. The classic Stroop tech- gesture and speech, one wonders what might happen in nique presents people with color words that are written the absence of explicit task demands. Indeed, if people in different colored fonts, and the “Stroop effect” arises strategically choose (on their own) not to integrate ges- when the meaning of the written word influences how ture and speech, it would allow for much stronger claims quickly and accurately people can name the color of about the inherent controlled nature of gesture–speech the font (Stroop, 1935). This effect has traditionally been integration. Second, the task required people to con- viewed as a hallmark for automatic processing (Logan, sciously attend to the semantics of the gesture–speech 1978; Schneider & Shiffrin, 1977; Posner & Snyder, 1975), utterances, which could have inherently encouraged con- but more recent research has shown that certain attentional trolled processing. A much more powerful test would be and contextual factors can modulate the effect (Bessner Kelly, Creigh, and Bartolotti 685
& Stolz, 1999; MacLeod, 1991). By using a variant of this turer and the speaker were the same person (a man or a Stroop procedure, we avoided the problems of explicitly woman), but in the gender-different condition, the ges- drawing attention to gesture and speech—attention that turer and the speaker were two different people (e.g., may unintentionally encourage conscious and strategic pro- a man gesturing but a woman speaking, and vice versa). cessing of the two modalities. The gender-same and gender-different conditions were In our modified version of the classic Stroop paradigm, created by digitally inserting the speech of the man or the participants watched videos of a man and a woman ges- woman into the video containing the gestures of the woman turing and speaking about common actions. The videos or the man. differed as to whether the gender of the gesturer and In addition to the gender relationship, the stimuli varied speaker was the same or different and whether the con- as to whether the gesture and the speech had a congruent tent of the gesture and speech was congruent or incon- or an incongruent relationship. In the gesture-congruent gruent. The task was to simply identify whether the man condition, the gesture and the speech were semantically or the woman produced the spoken portion of the vid- related (e.g., gesturing cut and saying “cut”), and in the eos while accuracy rates and RTs (behavioral measures) gesture-incongruent condition, they were semantically un- and ERPs (electrophysiological measure) were recorded to related (e.g., gesturing stir and saying “cut”). The congruent the spoken targets. In this way, participants were required gesture stimuli were normed before the experiment (n = 5) to make very superficial judgments about the acoustic to ensure that the words and the gestures were indeed se- properties of the spoken message, that is, whether a word mantically related. The norming procedure presented vi- was produced by a man or a woman. Importantly, the ac- deo clips to participants who had to rate (on a scale of 1 tual content of the gesture and speech was irrelevant. to 7) whether the speech and the gesture were related. Using this Stroop-like paradigm, we made two predic- There were initially 26 gesture–speech pairs, but because tions. The rationale of the predictions is that although two of them had ambiguous gesture–speech relationships, the relationship between gesture and speech was not rel- we dropped these two and were left with 24 pairs (that in- evant to the participantʼs task (i.e., to identify the gen- cludes one practice pair). The mean congruence rating of der of the spoken portion of the videos), it nevertheless the remaining 24 stimuli was 6.61 (SD = 0.63). We created should affect task performance if the integration of ges- the incongruent pairs by inserting the audio portion into ges- ture and speech is automatic. On the basis of this logic, ture videos that conveyed semantically unrelated informa- the first prediction is that participants should be slower tion. See Appendix A for all the gesture–speech pairs used to identify the gender of a speaker when gesture and as stimuli, and refer to Figure 1 for an example of the gender speech have an incongruent versus congruent relation- ship. The second prediction is that there will be a larger N400 component (indexing semantic integration difficul- ties) to words accompanied by incongruent versus congru- ent gestures. METHODS Participants Thirty-two right-handed (measured by the Edinburgh Handedness Inventory) Caucasian college undergraduates (12 males, 20 females; mean age = 20 years) participated for course credit. All participants signed an informed con- sent approved by the institutional review board. A total of three participants were not analyzed due to excessive artifacts in brain wave data, and one was removed for not following task instructions. Materials Participants watched digitized videos of a man and a woman (only the torso was visible) uttering common action verbs while producing iconic gestures also conveying actions. To test our two predictions, the videos were designed in a Stroop-like fashion. Specifically, there was a gender manip- ulation that varied the relationship between the gesturer Figure 1. Still frame examples of the congruent and incongruent and the speaker. In the gender-same condition, the ges- stimuli for the gender-same and gender-different videos. 686 Journal of Cognitive Neuroscience Volume 22, Number 4
and gesture conditions for the “cut” item. The stimuli were prestimulus window. Eye artifacts during data collection designed to create a Stroop-like test that required explicit at- were monitored with 4 EOG electrodes, with voltage shifts tention to the gender of the speaker, with the ultimate goal above 70 μV marked as bad (for more on the EOG algo- of determining whether people automatically process the se- rithm, see Miller, Gratton, & Yee, 1988; Gratton, Coles, & mantic relationship between speech and gesture (more Donchin, 1983). Non-EOG channels were marked as bad below). if there were shifts within the electrode of greater than Each audiovisual segment was 1 sec in duration. Differ- 200 μV for any single trial. If over 20% of the channels ent from previous research on the integration of speech were bad for a trial, the whole trial was rejected. In all, and cospeech gesture (cf. Kelly et al., 2007), the video 16% (SD = 22%) of the trials were rejected. began immediately with the stroke portion of the action The behavioral data were analyzed with a 2 (Gesture, gesture (i.e., there was no preparation phase as there was congruent and incongruent) × 2 (Gender, same and dif- in previous research). Following 200 msec after gesture ferent) repeated measures ANOVA with accuracy scores onset, the spoken portion of the video began and lasted and response latencies as the dependent measures. These for 800 msec. Throughout this 800 msec, the gesture con- behavioral responses were taken to the onset of the verbal tinued. In this way, gesture preceded speech for 200 msec potion of the videos. and overlapped with speech for 800 msec. All of the ERP The ERP data were analyzed with a 2 (Gesture, congru- data were time locked to speech onset.2 The variable ISI ent and incongruent) × 2 (Gender, same and different) × was 1.5 to 2.5 sec. The video was 25 min in length. 5 (central, frontal, occipital, parietal, and temporal elec- trode region) × 2 (left and right) repeated measures ANOVA. The present article focuses on one electrophysiological Procedure measure—the N400 component, which indexes semantic After participants were fitted with the ERP net (see be- integration (Kutas & Hillyard, 1980). The N400 window was low), the experimenter explained that participants would created by calculating the average amplitude from 250 to view videos of a man and a woman making short, one- 550 msec (N400) postspeech onset.3 word spoken utterances. The task was to identify, as quickly The electrode manipulation refers to the location of as possible, the gender of the verbal portion of the videos clusters of electrodes on the scalp. On the basis of pre- (i.e., whether the word was produced in a male or a fe- vious studies, the 128 electrodes were broken up into five male voice). A computer recorded these responses, and clusters of channels that corresponded roughly to basic their latencies were used in the RT analyses. In addition, anatomical structures of the brain. Refer to Figure 2 for the computer simultaneously recorded brain wave data a diagram of the clusters for the 128 electrodes (for more to the same responses. Note that this task does not re- on the clusters, see Kelly et al., 2004). The purpose of quire participants to respond to the semantic content of these electrode clusters was to test whether the N400 ef- the speech, nor does it require any attention to gesture. fect was stronger or weaker in particular scalp regions. Moreover, the gender of the gesturer is also irrelevant Specifically, the N400 effect is typically largest in central to the task. In this way, the task is a modified Stroop test, regions (Kutas & Hillyard, 1980) but is more anterior for with one piece of information relevant to the task—the stimuli that are comprised of gesture and speech (Kelly gender of the speaker—and other pieces of information et al., 2007; Wu & Coulson, 2007a). irrelevant to the task—the semantic relationship between All repeated measures analyses were adjusted for sphe- speech and gesture and the physical relationship between ricity by using a Greenhouse–Geisser correction (Howell, the speaker and the gesturer. 2002). Because all contrasts were planned and orthogo- This Stroop-like design allowed us to test our two pre- nal, Studentʼs t tests followed up on significant interac- dictions: if gesture and speech are automatically inte- tion effects. grated, participants should (1) be slower to identify the gender of a speaker when gesture and speech have an incongruent versus congruent relationship and (2) pro- RESULTS duce a larger N400 component to words accompanied by incongruent versus congruent gestures. Behavioral Data The ANOVA on the accuracy data revealed no significant main effects of Gesture, F(1,27) = 0.48, ns, or Gender, ERP Setup and Analysis F(1,27) = 2.03, ns. In addition, there was not a significant Participants were fitted with a 128-electrode Geodesic ERP interaction of Gesture × Gender, F(1,27) = 1.56, ns. net. The EEG was sampled at 250 Hz using a band-pass fil- These null results are likely due to a ceiling effect in the ter of 0.1–30 Hz, and impedances were kept below 40 kΩ accuracy scores (refer to Table 1). (the Geonet system uses high-impedance amplifiers). The The ANOVA on the RT data did not reveal a significant ERPs were vertex referenced for recording and linked- main effect of Gender, F(1,27) = 0.01 ns, but it did uncov- mastoid referenced for presentation. Following rereferenc- er a significant main effect of Gesture, F(1,27) = 32.60, ing, the brain waves were baseline corrected to a 100-msec p < .001. In addition, there was a significant interaction of Kelly, Creigh, and Bartolotti 687
Figure 2. Ten electrode clusters for the 128-electrode geodesic net. For more on the rationale for the clusters, see Kelly et al. (2004). Gesture × Gender, F(1,27) = 9.04, p = .006. Simple effects ent study, there was a significant main effect of Gender, analyses demonstrated that within the gender-same con- F(1,27) = 4.76, p = .038, with gender-different stimuli dition, participants were slower when gestures conveyed producing a greater negativity than gender-same stim- incongruent information to speech (M = 818 msec, SD = uli.4 In addition, in accordance with the predictions, there 198 msec) compared with when it conveyed congruent in- was a significant main effect of Gesture, F(1,27) = 6.95, formation (M = 772 msec, SD = 207 msec), t(27) = 5.34, p = .013, with gesture-incongruent stimuli producing a p < .001; and within the gender-different condition, par- more negative going N400 component than gesture- ticipants were also slower—but to a lesser degree—when congruent stimuli. Refer to Figure 3 for the main effect gesture and speech were incongruent (M = 805 msec, of Gesture, Figure 4 for the effects at different electrode SD = 198 msec) versus congruent (M = 786 msec, SD = sites, and Figure 5 for a topographical map of the N400 208 msec), t(27) = 3.39, p < .001 (refer to Table 1). Note effect (peak = 408 msec) across all electrode sites on that although the gesture-incongruent condition was slower the scalp. Beyond these main effects, there were no sig- than the gesture-congruent condition in both gender con- nificant two-way interactions for Gesture × Electrode Re- ditions, the significant interaction reveals that this dif- gion, F(4,108) = 1.23, ns, Gender × Electrode Region, ference was greater when the gender of the gesturer and F(4,108) = 2.62, ns, Gesture × Hemisphere, F(1,27) = speaker was the same versus different. 1.02, ns, Gender × Hemisphere, F(1,27) = 0.12, ns, or These results confirm our first prediction: Participants Gesture × Gender, F(1,27) = 0.01, ns. Nor were there were slower to identify the gender of the speaker when any three- or four-way interactions involving Gesture and gesture and speech had an incongruent versus congruent Gender variables. relationship, even when the relationship between the From observation of Figures 3 and 4, it appears that two modalities was not relevant to the task. the difference between the gesture-congruent and the gesture-incongruent conditions may be the result of a carry over from differences earlier at the P2 peak.5 To ad- Electrophysiological Data dress this possibility, we subtracted the P2 window (150– The ANOVA on the N400 time window revealed two main 250 msec) from the N400 window and reran the ANOVA. effects. Although not part of the predictions of the pres- Even after using this new baseline, the gesture-incongruent Table 1. Accuracy Scores and RTs across the Gesture and Gender Conditions Mean Proportion Correct (SD) Mean RT (SD) Gesture Congruent Gesture Incongruent Gesture Congruent (msec) Gesture Incongruent (msec) Gender same 0.99 (0.03) 0.98 (0.03) 772 (207) 818 (208) Gender different 0.97 (0.03) 0.98 (0.03) 786 (198) 805 (199) 688 Journal of Cognitive Neuroscience Volume 22, Number 4
Figure 3. Grand averaged brain waves (collapsed across all electrode regions) for the gesture main effect. The gesture-incongruent condition produced a larger N400 component than the gesture-congruent condition. Microvolts are plotted negative up. stimuli produced a more negative going N400 component the gesture-incongruent stimuli producing a larger neg- than the gesture-congruent stimuli, F(1,27) = 4.99, p = ativity than gesture-congruent stimuli in bilateral Central, .034. Interestingly, this new analysis also yielded a mar- F(1,27) = 5.88, p = .022, and Parietal regions, F(1,27) = ginally significant interaction of Gesture × Electrode re- 8.40, p = .007. However, there was not a significant effect gion, F(4,108) = 2.49, p = .103. This effect was driven by in bilateral frontal regions, F(1,27) = 1.26, ns. Figure 4. Topographical plots for the 10 electrode regions for the gesture condition. The significant N400 effects are boxed in bilateral central and parietal electrode regions. Microvolts are plotted negative up. Kelly, Creigh, and Bartolotti 689
Figure 5. Topographical maps for the N400 peaks (408 msec) for the two gesture conditions. The top of each map is toward the front of the head. Hot colors are positive, and cool colors are negative. The figure can be viewed in color on line. These results from the electrophysiological measure gesture–speech stimuli, even when the semantic relation- confirm our second prediction: Although the relationship ship of gesture and speech had nothing to do with the task between gesture and speech was irrelevant to the partic- at hand. ipantsʼ task, gestures that were semantically incongruent Regarding the other two hallmarks—low level and fast— to the content of the speakerʼs words produced a larger our results are less conclusive, but they are at least sugges- N400 component than gestures that were semantically tive of automatic processing. In some recent neuroimaging congruent. research on the neural mechanisms of gesture–speech integration, Hubbard et al. (2009) argued that the planum temporale integrates the prosodic elements of speech and DISCUSSION gesture (specifically beat gestures that convey no iconic The results of the experiment support our two predic- content), suggesting that these types of gesture influence tions: Participants were slower to process the gender of low-level acoustic analysis of spoken input. More generally, the speakerʼs voice when the content of gesture and Iacoboni et al. (2005) have argued that low-level and phylo- speech—information not relevant to the task—was in- genetically old neural mechanisms (e.g., the mirror neuron congruent versus congruent (Prediction 1), and partici- system, which overlaps with the planum temporale) are pants produced a larger N400 component when gesture designed to automatically process manual hand actions. and speech were incongruent versus congruent (Predic- Considered in this context, one can speculate that iconic tion 2). These findings are consistent with the claim that hand gestures may interact with low-level, perceptual as- gesture and speech are automatically integrated during pects of speech processing in a similarly automatic fashion. language comprehension. When it comes to the speed of gesture–speech integra- tion, the present results are consistent with other ERP stud- ies on gesture–speech processing that have identified Gesture–Speech Integration as an early semantic components (Bernardis et al., 2008; Holle Automatic Process & Gunter, 2007; Özyürek et al., 2007; Wu & Coulson, The traditional hallmarks of automatic processing are that 2007a, 2007b). Although we did not report ERP effects it is low level, fast, and obligatory (Logan, 1978; Schneider earlier than the N400 effect, previous research has uncov- & Shiffrin, 1977; Posner & Snyder, 1975). Using a modi- ered evidence for even earlier integration of verbal and fied Stroop paradigm, we provided definitive evidence nonverbal information ( Willems, Özyürek, & Hagoort, for the third aspect of automaticity—the obligatory nature 2008; Kelly et al., 2004). In the Kelly study, the presence of gesture–speech processing. Our procedure did not re- of hand gesture influenced the P2 component and even quire any overt attention to the semantic content of speech earlier sensory stages of speech processing (within the and gesture, yet the content of gesture influenced the first 150 msec). Although the present article did not ex- superficial acoustic analysis of speech in both the behav- plore such early effects, the two studies together suggest ioral and the electrophysiological measures. That is, when that gesture and speech may be automatically integrated participants simply had to process the auditory differences very quickly in the brain. between a male and a female voice—differences that were In general, this claim about the automaticity of gesture extremely salient—they could not do so without also pro- processing is consistent with theories that the human cessing irrelevant information conveyed through gesture brain has evolved specialized neural mechanisms to auto- and speech. This interference resulted in slower RTs and matically process nonverbal behaviors, such as bodily ges- a larger N400 component to incongruent versus congruent tures, facial expressions, and manual actions (de Gelder, 690 Journal of Cognitive Neuroscience Volume 22, Number 4
2006; Iacoboni et al., 2005; Blakemore & Decety, 2001). which meaningful gestures are integrated with speech dur- Building on this, if hand gestures served as a foundation ing sentence processing. Moreover, Kelly et al. (2007) ar- for the evolution of spoken language, as some have ar- gued that explicit knowledge about a communicatorʼs gued (Corballis, 2003; Bates & Dick, 2002; Kelly et al., intent modulates the neural integration of gesture and 2002; Rizzolatti & Arbib, 1998), it makes sense that these speech during word comprehension. Similarly, it is possi- hand movements would be automatically linked to ble that on some level, participants in the present study speech processes. Thus, in contrast to views that gesture were aware that gesture and speech produced by different is a mere afterthought in language processing, gesture people did not “belong” together, and consequently, they appears to be a real contender in communication, one integrated the two modalities differently than when ges- that has an inextricable link with the speech that it natu- ture and speech did belong together. rally and ubiquitously accompanies. This connection to previous research bears further dis- Importantly, the present study extends previous re- cussion. In the study by Kelly et al. (2007), there were very search on the automatic integration of gesture and speech similar patterns of behavioral data—in both studies, partic- (Kelly et al., 2007). In this previous work, participants were ipants integrated gesture and speech to a greater extent explicitly instructed to attend to the meaning of speech. In when the two modalities were meant to go together. How- contrast, by using a Stroop-like paradigm in the present ever, unlike that previous research, the present study did study, participants were not required to attend to the not find a comparable pattern with the electrophysiologi- meaning of speech—or gesture—but did so anyway. More cal data. Specifically, in the previous work, the N400 effect importantly, the present results shed light on the role of for congruent and incongruent gesture–speech pairs was iconicity in the integration of gesture and speech. Unlike eliminated in left hemisphere regions when participants previous research that conflated indexicality and iconicity, believed that the gesture and speech were not intention- the present results demonstrated that the representational ally coupled (the no-intent condition). However, in the content (i.e., the iconic meaning) of gesture is automati- present study, there was a strong bilateral N400 effect cally connected to the representational content of speech. for congruent and incongruent gesture–speech pairs even This finding strongly supports theories that speech and when gesture and speech did not belong together (the iconic gesture—at least iconic action gestures—are inte- gender-different condition). One possible explanation for grated on a deep conceptual level (McNeill, 1992, 2005; this difference is that in previous research, participants Kita & Özyürek, 2003). To further explore this relation- were explicitly told that gesture and speech did not belong ship, it would be interesting for future research to in- together, whereas in the present study, they had to de- vestigate the automatic integration of speech and iconic termine this for themselves. In fact, participants may have nonaction gestures (e.g., gestures about object attributes assumed that gesture and speech were meant to go to- or spatial relationships) that have an even more ambigu- gether in the present study. After all, people have experi- ous meaning in the absence of accompanying speech. ence with video media that artificially combine language and action (e.g., as in watching a dubbed movie or docu- mentary footage with a narrator). Perhaps participants ap- Gesture–Speech Integration Is Not an Exclusively plied this sort of previous experience to the stimuli used Automatic Process in the present study, and this is why the ERP data did not Although we found strong support for the obligatory na- correspond to the previous research. In any event, it will ture of gesture–speech integration, the RT data suggest be important for future research to determine the condi- that it may not be an exclusively automatic process. Recall tions under which people naturally integrate gesture and that there were larger RT differences between the two speech and when this integration is modulated by higher gesture–speech pairs in the gender-same versus gender- level contextual information. different condition. This suggests that participants were Focusing just on the present study, it is important to ad- sensitive to the context within which they processed ges- dress another inconsistency—why did the gender context ture and speech. That is, when the context encouraged modulate the behavioral but not the electrophysiological integration of gesture and speech (i.e., when the same integration of gesture and speech? This sort of inconsis- person produced the gesture and word), participants were tency between the behavioral and the electrophysiological more likely to combine the two modalities than when the findings is actually consistent with previous ERP research context did not encourage integration (i.e., when different on the N400 effect (Ruz, Madrid, Lupiánez, & Tudela, people produced the gesture and word). This finding fits 2003; Brown & Hagoort, 1993; Holcomb, 1993). For exam- with claims that the Stroop effect can be modulated by ple, Brown and Hagoort (1993) found that incongruent higher level contextual variables (Bessner & Stolz, 1999). masked primes produced slower RTs to targets than con- Moreover, this modulation is consistent with previous gruent masked primes, but this effect was not carried over research on the controlled nature of gesture–speech in- in the electrophysiological data—there was no corre- tegration (Holle & Gunter, 2007; Kelly et al., 2007). For sponding N400 effect for masked incongruent items. This example, Holle and Gunter (2007) showed that the pres- suggests that behavioral measures can occasionally reveal ence of nonmeaningful gestures influences the extent to differences that electrophysiological measures cannot. Kelly, Creigh, and Bartolotti 691
Although this inconsistency tempers strong claims that ioral measure) that people are sensitive to whether speech context modulates the integration of gesture and speech, and gesture “belong” together, suggesting a potential con- it encourages future work to explore the circumstances trolled mechanism as well. Although it will be important to under which controlled mechanisms play a definitive role build on these findings using more natural stimuli in larger in the processing of gesture and speech during language discourse contexts, they serve as a preliminary foundation comprehension. for the claim that gesture and speech are inextricably linked during language comprehension. And more gener- ally, by viewing gesture as a unique interface between Local versus Global Integration actions and words, this work will hopefully lend a hand Previous research has made a distinction between “local” to the growing research on the relationship between body and “global” integration of gesture and speech (Özyürek and language in the human brain. et al., 2007). Local integration concerns how a gesture is integrated with a temporally overlapping word (i.e., when a gesture and a word are presented at roughly the same time). In contrast, global integration concerns how a APPENDIX A gesture is integrated with speech over larger spans of discourse (e.g., when a gesture precedes a word in a List of the 23 Congruent and Incongruent sentence or vice versa). The majority of ERP research on Gesture–Speech Pairs gesture–speech integration has found evidence for glob- Congruent Incongruent al integration (Bernardis et al., 2008; Holle & Gunter, 2007; Özyürek et al., 2007; Wu & Coulson, 2007a, 2007b), Speech Gesture Speech Gesture but the present study has provided evidence for local Beat Beat Beat Wipe integration. On the surface, this finding is inconsistent with the work Brush Brush Brush Shake by Özyürek et al. (2007) who, using gestures in larger sen- Chop Chop Chop Mow tence contexts, found evidence for global—but not local— integration. One explanation for this inconsistency is that Close Close Close Screw there were different SOAs (the timing difference between Cut Cut Cut Stir the onset of speech and gesture) when measuring the local Hammer Hammer Hammer Twist integration of gesture and speech in the two studies. Whereas the present study found an N400 effect for incon- Knock Knock Knock Write gruent gesture–speech pairs using SOAs of 200 msec, the Lift Lift Lift Stab work by Özyürek et al. had a simultaneous onset of ges- Mow Mow Mow Turn ture and speech and they found no N400 effect for local gesture–speech incongruencies. One possible explanation Roll Roll Roll Hammer for this inconsistency is that local integration occurs only Saw Saw Saw Type for gesture–speech pairs that are removed from natural discourse (as in the present study). An equally plausible Screw Screw Screw Cut possibility is that gesture and speech are indeed locally in- Shake Shake Shake Close tegrated in natural discourse, but not when they are simul- Squeeze Squeeze Squeeze Lift taneously presented (as in the study by Özyürek et al., 2007). Indeed, gesture slightly precedes speech onset in Stab Stab Stab Sweep natural discourse (McNeill, 1992), and presenting them Stir Stir Stir Break simultaneously may disrupt their natural integration. It will Sweep Sweep Sweep Tear be important for future ERP research to directly compare different gesture–speech SOAs to more thoroughly under- Tear Tear Tear Roll stand how the timing of gesture and speech affects the Turn Turn Turn Chop neural integration of the two modalities. Twist Twist Twist Knock Type Type Type Brush Conclusion Wipe Wipe Wipe Squeeze By using a variation on the Stroop procedure, we demon- strated that people obligatorily integrate speech and iconic Write Write Write Saw gestures even when that integration is not an explicit re- quirement. This suggests that the neural processing of Practice speech and iconic gesture is to some extent automatic. Wring Wring Wring Scrub However, there is also evidence (at least with the behav- 692 Journal of Cognitive Neuroscience Volume 22, Number 4
Acknowledgments Bessner, D., & Stolz, J. A. (1999). Unconsciously controlled processing: The Stroop effect reconsidered. Psychonomic We would like to thank the anonymous reviewers for their help- Bulletin & Review, 6, 99–104. ful suggestions on previous versions of this manuscript. In ad- Blakemore, S. J., & Decety, J. (2001). From the perception of dition, we are grateful to Ron Crans for technical assistance in action to the understanding of intention. Nature Reviews software creation. Finally, we thank the Colgate University for Neuroscience, 2, 561–567. providing generous support for undergraduate research. Brown, C., & Hagoort, P. (1993). The processing nature of Reprint requests should be sent to Spencer Kelly, Department the N400: Evidence from masked priming. Journal of of Psychology, Neuroscience Program, Colgate University, 13 Cognitive Neuroscience, 5, 34–44. Oak Dr., Hamilton, NY 13346, or via e-mail: skelly@colgate.edu. Cassell, J., McNeill, D., & McCullough, K. E. (1999). Speech–gesture mismatches: Evidence for one underlying representation of linguistic and nonlinguistic information. Pragmatics & Cognition, 7, 1–34. Notes Clark, H. H. (1996). Using language. Cambridge, England: Cambridge University Press. 1. This, of course, is not to say that these are the only two ERP Corballis, M. C. (2003). From hand to mouth: The origins of studies to find an N400 effect using speech and gesture stimuli language. Princeton, NJ: Princeton University Press. (see Bernardis et al., 2008; Holle & Gunter, 2007; Özyürek et al., de Gelder, B. (2006). Towards the neurobiology of 2007; Wu & Coulson, 2007a, 2007b). However, they are the only emotional body language. Nature Reviews Neuroscience, two studies to find an N400 effect to speech stimuli that were simul- 7, 242–249. taneously coupled with incongruent versus congruent gestures. Goldin-Meadow, S. (2003). Hearing gesture: How our hands 2. Because the gestures preceded speech by such a short help us think. Cambridge, MA: Belknap Press. interval (200 msec) and because the ERPs were time locked to Goldin-Meadow, S., Wein, D., & Chang, C. (1992). Assessing speech that co-occurred with gesture, the brainwaves had a knowledge through gesture: Using childrenʼs hands positive skew, especially in frontal regions. This skew is common to read their minds. Cognition and Instruction, 9, in priming studies that employ very short SOAs (e.g., Kiefer & 201–219. Brendel, 2006), and the positive slow wave most likely reflects Gratton, G., Coles, M. G. H., & Donchin, E. (1983). A new the electrophysiological response to speech riding on top of method for off-line removal of ocular artifacts. an electrophysiological response to the gesture. Importantly, this Electroencephalography and Clinical Neurophysiology, positive skew is uniform for all stimuli, so it should not confound 55, 468–484. any of our treatment conditions. Holcomb, P. (1993). Semantic priming and stimulus 3. Note that 250 msec is relatively early for the start of the N400 degradation: Implications for the role of the N400 in component, but there is precedence in the literature for this early language processing. Psychophysiology, 30, 47–61. onset (e.g., Van Berkum, van den Brink, Tesink, Kos, & Hagoort, Holle, H., & Gunter, T. C. (2007). The role of iconic gestures in 2008). A post hoc analysis showed that the early negativity was speech disambiguation: ERP evidence. Journal of Cognitive not a separate N300 component (McPherson & Holcomb, 1999). Neuroscience, 19, 1175–1192. 4. Although not central to the present study, this gender Holle, H., Gunter, T., Rüschemeyer, S. A., Hennenlotter, A., & effect is noteworthy because it is related to a recent ERP study Iacoboni, M. (2008). Neural correlates of the processing of demonstrating a large N400 component when people hear co-speech gestures. Neuroimage, 39, 2010–2024. utterances that are unusual for a speaker (e.g., a man saying Howell, D. C. (2002). Statistical methods for psychology that he wished he looked like Britney Spears; Van Berkum et al., (5th ed.). Belmont, CA: Duxberry Press. 2008). The present findings suggest that simply seeing the gen- Hubbard, A. L., Wilson, S. M., Callan, D. E., & Dapretto, M. der of speaker sets up a “semantic” expectation of the gender of (2009). Giving speech a hand: Gesture modulates activity in the voice it accompanies. auditory cortex during speech perception. Human Brain 5. We do not explore the P2 effect in the present manuscript. Mapping, 30, 1028–1037. For coverage of the early integration of gesture and speech, see Iacoboni, M., Molnar-Szakacs, I., Gallese, V., Buccino, G., Kelly et al., 2004. Mazziotta, J. C., & Rizzolatti, G. (2005). Grasping intentions of others with oneʼs own mirror neuron system. PloS Biology, 3, 529–535. REFERENCES Kelly, S. D., Barr, D., Church, R. B., & Lynch, K. (1999). Offering a hand to pragmatic understanding: The role of speech Armstrong, D. F., & Wilcox, S. E. (2007). The gestural origin of and gesture in comprehension and memory. Journal of language. New York: Oxford University Press. Memory and Language, 40, 577–592. Bargh, J. A., & Morsella, E. (2008). The unconscious mind. Kelly, S. D., & Church, R. B. (1998). A comparison Perspectives on Psychological Science, 3, 73–79. between childrenʼs and adultsʼ ability to detect childrenʼs Bates, E., Benigni, L., Bretherton, I., Camaioni, L., & Volterra, V. representational gestures. Child Development, 69, (1979). The emergence of symbols: Cognition and 85–93. communication in infancy. New York: Academic Press. Kelly, S. D., Iverson, J., Terranova, J., Niego, J., Hopkins, M., & Bates, E., & Dick, F. (2002). Language, gesture, and the Goldsmith, L. (2002). Putting language back in the body: developing brain. Developmental Psychobiology, 40, 293–310. Speech and gesture on three timeframes. Developmental Beattie, G., & Shovelton, H. (1999). Do iconic hand gestures Neuropsychology, 22, 323–349. really contribute anything to the semantic information Kelly, S. D., Kravitz, C., & Hopkins, M. (2004). Neural correlates conveyed by speech? An experimental investigation. of bimodal speech and gesture comprehension. Brain and Semiotica, 123, 1–30. Language, 89, 253–260. Bernardis, P., Salillas, E., & Caramelli, N. (2008). Behavioural Kelly, S. D., Ward, S., Creigh, P., & Bartolotti, J. (2007). An and neurophysiological evidence of semantic interaction intentional stance modulates the integration of gesture and between iconic gestures and words. Cognitive speech during comprehension. Brain and Language, 101, Neuropsychology, 25, 1114–1128. 222–233. Kelly, Creigh, and Bartolotti 693
Kendon, A. (2004). Gesture: Visible action as utterance. and gesture: Insights from event-related brain potentials. Cambridge: Cambridge University Press. Journal of Cognitive Neuroscience, 19, 605–616. Kiefer, M., & Brendel, D. (2006). Attentional modulation Posner, M. I., & Snyder, C. R. R. (1975). Attention and cognitive of unconscious “automatic” processes: Evidence from control. In R. L. Solso (Ed.), Information processing and event-related potentials in a masked priming paradigm. cognition: The Loyola symposium (pp. 55–85). Hillsdale, NJ: Journal of Cognitive Neuroscience, 18, 184–198. Lawrence Erlbaum. Kita, S., & Özyürek, A. (2003). What does cross-linguistic Pourtois, G., de Gelder, B., Vroomen, J., Rossion, B., & variation in semantic coordination of speech and gesture Crommelinck, M. (2000). The time-course of intermodal reveal?: Evidence for an interface representation of spatial binding between hearing and seeing affective information. thinking and speaking. Journal of Memory and Language, NeuroReport, 11, 1329–1333. 48, 16–32. Rizzolatti, G., & Arbib, M. A. (1998). Language within our grasp. Krauss, R. M. (1998). Why do we gesture when we speak? Trends in Neurosciences, 21, 188–194. Current Directions in Psychological Science, 7, 54–59. Rizzolatti, G., & Craighero, L. (2004). The mirror-neuron Krauss, R. M., Morrel-Samuels, P., & Colasante, C. (1991). Do system. Annual Review of Neuroscience, 27, 169–192. conversational hand gestures communicate? Journal of Ruz, M., Madrid, E., Lupiánez, J., & Tudela, P. (2003). High Personality and Social Psychology, 61, 743–754. density ERP indices of conscious and unconscious Kutas, M., & Hillyard, S. A. (1980). Reading senseless sentences: semantic priming. Cognitive Brain Research, 17, Brain potentials reflect semantic incongruity. Science, 207, 719–731. 203–204. Schneider, W., & Shiffrin, R. M. (1977). Controlled and Langton, S. R. H., & Bruce, V. (2000). You must see the point: automatic human information processing: Detection, search, Automatic processing of cues to the direction of social and attention. Psychological Review, 84, 1–66. attention. Journal of Experimental Psychology: Human Skipper, J. I., Goldin-Meadow, S., Nusbaum, H. C., & Small, Perception and Performance, 26, 747–757. S. L. (2007). Speech-associated gestures, Brocaʼs area, and Logan, G. D. (1978). Attention in character classification: the human mirror system. Brain and Language, 101, Evidence for the automaticity of component stages. 260–277. Journal of Experimental Psychology: General, 107, Stroop, J. R. (1935). Studies of interference in serial verbal 32–63. reactions. Journal of Experimental Psychology, 18, 643–662. MacLeod, C. M. (1991). Half a century of research on the Stroop Van Berkum, J. J. J., van den Brink, D., Tesink, C., Kos, M., effect: An integrative review. Psychological Bulletin, 109, & Hagoort, P. (2008). The neural integration of speaker 163–203. and message. Journal of Cognitive Neuroscience, 20, MacWhinney, B., & Bates, E. (1989). The crosslinguistic 580–591. study of sentence processing. New York: Cambridge Willems, R. M., & Hagoort, P. (2007). Neural evidence for the University Press. interplay between language, gesture, and action: A review. McNeill, D. (1992). Hand and mind: What gesture reveals Brain and Language, 101, 278–289. about thoughts. Chicago: University of Chicago Press. Willems, R. M., Özyürek, A., & Hagoort, P. (2007). When McNeill, D. (2005). Gesture and thought. Chicago: University language meets action: The neural integration of gesture and of Chicago Press. speech. Cerebral Cortex, 17, 2322–2333. McPherson, W. B., & Holcomb, P. J. (1999). An Willems, R. M., Özyürek, A., & Hagoort, P. (2008). Seeing and electrophysiological investigation of semantic priming hearing meaning: ERP and fMRI evidence of word versus with pictures of real objects. Psychophysiology, 36, picture integration into a sentence context. Journal of 53–65. Cognitive Neuroscience, 20, 1235–1249. Miller, G. A., Gratton, G., & Yee, C. M. (1988). Generalized Wilson, S. M., Molnar-Szakacs, I., & Iacoboni, M. (2008). implementation of an eye movement correction procedure. Beyond superior temporal cortex: Intersubject correlations Psychophysiology, 25, 241–243. in narrative speech comprehension. Cerebral Cortex, 18, Montgomery, K. J., Isenberg, N., & Haxby, J. V. (2007). 230–242. Communicative hand gestures and object-directed hand Winston, J. S., Strange, B. A., OʼDoherty, J., & Dolan, R. J. movements activated the mirror neuron system. Social (2002). Automatic and intentional brain responses during Cognitive & Affective Neuroscience, 2, 114–122. evaluation of trustworthiness of faces. Nature Neuroscience, Nishitani, N., Schurmann, M., Amunts, K., & Hari, R. (2005). 5, 277–283. Brocaʼs region: From action to language. Physiology, 20, Wu, Y. C., & Coulson, S. (2007a). Iconic gestures prime related 60–69. concepts: An ERP study. Psychonomic Bulletin & Review, 14, Özyürek, A., & Kelly, S. D. (2007). Gesture, brain, and language. 57–63. Brain and Language, 101, 181–184. Wu, Y. C., & Coulson, S. (2007b). How iconic gestures enhance Özyürek, A., Willems, R. M., Kita, S., & Hagoort, P. (2007). communication: An ERP study. Brain and Language, 101, On-line integration of semantic information from speech 234–245. 694 Journal of Cognitive Neuroscience Volume 22, Number 4
You can also read